| Hi-index | 0.00 |
The representation and understanding of physical objects unify diverse fields in and around Artificial Intelligence Research. They bring together the heuristic and epistemological categories. Although much of concern in the past has been on the former, the recent flood of literature indicates the growing interest in the latter.The components of a physical object configuration are often characterized by uncertainty. They arise often due to unreliability of the sources of information and incomplete information, as in the case of recognition of partially occluded scenes. Zadeh and others have advocated a theory of possibility to consider propositions with uncertainty. It appears promising to extend some of these ideas to propositions describing physical object configurations.Further, a scene could be described from multiple perspectives resulting in corresponding descriptions. It is often necessary to aggregate these propositions to higher levels of abstraction.The proposition describing stability must take into account the shape, size, and relative orientation of objects among themselves and with respect to supporting structures. In the case of vertical structures the perception of stability is a function of height and base. Often the third dimension is not easily fathomed, though some default assignment could be made. On consulting many observers the authors found that in the case of a vertical rectangular structure of width W and height H, a stability indicator S could be defined as S= 2(2W/H)2 for 02 for 0.5When objects are stacked one on top of another, it becomes necessary to combine the stability indicators to obtain the overall stability indicator for the configuration. This is done in the lines of combining evidences in the expert systems. However the Boolean operators AND, OR and NOT even with modified semantics do not seem to have an obvious parallel in spacial situations. In the case of a rectangular beam supported by two pillars the stability indicator S for the composite system could be obtained from S=1-(1-s1)(1-s2) where s1 and s2 are the indicators for the supports. Expressions for combining evidences obtained from multiple views of the same configurations are formulated by experiment and trial. Using such indicators simple rules on different kinds of configurations are developed. The high level reasoner for handling complex scenes is a Lisp based system that operates on propositions based on the indicators with a flexible control strategy.We are conceptualizing the stability of the configurations by appealing to commonsense judgements, instead of appealing to the laws of physics involving the determination of centres of gravity and computation of various forces. Such an approach would lack the rigor and reliability of the traditional methods. The advantages arise from the simplicity of computation and approximates the way humans rationalize on physical objects. In a robot task planning it is often necessary to avoid complicated computations and obtain working conclusions based on fewer and simplified parameters.